Always-on taste curation — the free moodboard extension for creatives
Get the free extension
PortfolioPricingTasteBlog
Back to Hub

Meta's Andromeda Engine: Why Micro-Testing is Dead in 2026

Meta's Andromeda Engine: Why Micro-Testing is Dead in 2026

How the shift from Ad IDs to Entity IDs in the pre-auction retrieval phase forces a complete rethink of creative testing.

G
Growth Engineering
Creative OperationsPublished 2026-08-02Updated May 1, 2026

Why is my new Meta ad not getting any spend?

In 2026, Meta's 'Andromeda' retrieval engine filters ads before they enter the auction. It identifies ads by their underlying 'Entity ID' (the core visual concept) rather than the unique 'Ad ID'. If your new ad is merely a micro-iteration (e.g., a button color change) of an existing ad, Andromeda classifies it as the same Entity ID. To prevent auction overlap, it suppresses the new variant, allocating all spend to the historical winner.

For years, media buying agencies built entire retainers around "micro-testing." They would take a winning video, change the background color of the CTA button, upload it as a new ad, and charge for creative iteration.

In 2026, this workflow is not just inefficient; it is actively penalized by the algorithm. The introduction of the Andromeda retrieval system fundamentally changed how Meta decides which ads participate in the auction.

Entity ID vs. Ad ID

Historically, every time you uploaded a file, Meta generated a unique Ad ID. The auction treated each Ad ID as an independent competitor.

Today, Meta's computer vision algorithms analyze the pixels, audio, and pacing of your upload to generate an Entity ID. The Entity ID represents the concept. If you upload a video where the only difference is the color of the text overlay, Meta assigns both videos the same Entity ID.

Testing MethodologyExampleAndromeda ClassificationResult
Micro-IterationRed CTA vs Blue CTASame Entity IDNew ad receives $0 spend.
Minor Hook SwapActor smiling vs Actor pointingOften Same Entity IDNew ad struggles for initial delivery.
Concept Testing (Ad Families)Founder Story vs UGC UnboxingDistinct Entity IDsBoth ads enter auction; data is generated.

Status

Optimal

The Pre-Auction Filter

  • Primary Filter MechanicConcept Diversity
  • Micro-Test Success Rate< 5%

Recommendation:Stop A/B testing minor variables. Shift your creative pipeline to generate entirely distinct Ad Families (different actors, different visual pacing, different emotional framing) to force the algorithm to evaluate them independently.

How the suppression works: Before the GEM auction calculates the winning bid, Andromeda retrieves a candidate pool of ads. To maximize user experience, Andromeda enforces diversity. If you have two ads with the same Entity ID, it drops the unproven variant from the candidate pool entirely. It never even reaches the bidding phase.

The Programmatic Shift

Because Andromeda demands distinct conceptual diversity, creative teams must dramatically increase the variance of their inputs. Hand-editing one or two variants in CapCut is mathematically insufficient to trigger diverse Entity IDs.

Media buyers are moving to programmatic assembly tools like eonikto solve this. Instead of slightly tweaking one video, they use eonik to programmatically merge distinct, diverse AI-generated hooks (wildly different visual pattern interrupts) with various B-roll bodies. This automated workflow generates the high-variance "Ad Families" required to bypass the Andromeda filter and secure actual spend in the auction.

Related Essays

EssayJuly 25, 2026

The End of the Agency Retainer

Why the era of paying $15,000 a month for 30 video variations is over, and how programmatic assembly is shifting the balance of power back to the brand.

Read Essay
EssayJuly 25, 2026

Why AI Editing Fails Without Human Strategy

You can generate 1,000 video variations a minute, but if the foundational psychology is wrong, you just created 1,000 losing ads. Here is why the "human-in-the-loop" is mandatory.

Read Essay

What to read next

Continue with guides that match where you are — production, research, methodology, or shortlist.
  • Paid social creative blog hub

    Read guide
  • How to generate AI ads

    how to generate ai ads workflow

    Read guide
  • Technical playbooks

    paid social implementation guides hub

    Read guide
  • On brand AI ad workflow

    Read guide
  • Creative testing methodology

    creative testing methodology team ops

    Read guide
  • Ad variant testing workflow

    ad variant testing workflow steps

    Read guide

The Mac app for making ads

Your next ad, without the busywork.

Bring your footage and your AI clips. eonik puts together the finished, on-brand cut — and you approve every frame before it ships.

macOS 15+ · Apple Silicon · Free to start

1
eonik

Finished, on-brand ads without the busywork.

Product

  • Pricing
  • Creative Testing
  • MCP for agents

Knowledge

  • Ad Library
  • Knowledge Hub
  • Blog

Solutions

  • DTC Brands
  • Agencies
  • Growth Teams

Company

  • About
  • Community
  • connect@eonik.ai
PrivacyTerms

© 2026 eonik. All rights reserved.